Re: not translating backend types for ps and log_line_prefix

On Wed, Dec 10, 2025, at 6:41 PM, Álvaro Herrera wrote:
> On 2025-Dec-10, Heikki Linnakangas wrote:
>
>> > +PG_PROCTYPE(B_AUTOVAC_LAUNCHER, "autovacuum", gettext_noop("autovacuum launcher"), AutoVacLauncherMain, true)
>> > +PG_PROCTYPE(B_AUTOVAC_WORKER, "autovacuum", gettext_noop("autovacuum worker"), AutoVacWorkerMain, true)
>>
>> It seems not nice that both have the same name, "autovacuum". Similarly, I
>> think it's good to have different names for dead-end backends and regular
>> backends.
>
> True, I was not sure about this either. For context, Euler's patch[2] was
> adding a new column called "category" for use with his new
> representation of log_min_messages, and he decided that something like
> "autovacuum:DEBUG" should affect both the worker and the launcher. I'm
> not sure myself that this is a good idea, because if you want DEBUG for
> just one and not the other, the proposed interface doesn't let you do
> that.
>

It is called "category" to group process types. The main motivation is
"backend" that has multiple entries. For "autovacuum", I checked the code and
launcher has just a few messages so I decided to have just one category. That's
certainly not a strong argument.

> [2] https://postgr.es/m/144f0f96-c6d3-4ad6-a236-00200513e2e0@app.fastmail.com
>
> I'd be okay with instead adding "avlauncher", "avworker", "dead-end",
> "standalone", to differentiate all those types. ("av" sounds perhaps
> too mysterious, so maybe "autovaclauncher" and "autovacworker" for
> those?) Bikeshedding, again, welcome.
>

I'm fine expanding the list. If we carefully choose the names we can certainly
expand it later. For example, the standalone can be a new category in the
future. The same can be applied to autovacuum ("autovacuum" controls the worker
and a new category "autovacumlauncher" controls launcher messages).

--
Euler Taveira
EDB https://www.enterprisedb.com/